Skip to content

fix: Remove CHECK_CONTIGUOUS for SFA/SFB #1963

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
+0 −5
Open

fix: Remove CHECK_CONTIGUOUS for SFA/SFB #1963

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 0 additions & 5 deletions csrc/gemm_groupwise_sm120.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,11 +89,6 @@ void CutlassGemmGroupwiseScaledSM120(TensorView float_workspace_buffer, TensorVi
cudaSetDevice(float_workspace_buffer->device.device_id);
auto stream = get_stream(C->device);

// Ensure scales are contiguous
// Note: We keep the original shape and let the kernel's layout handle interpretation
CHECK_CONTIGUOUS(SFA);
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do we have any assumptions on the layout of SFA or SFB?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The layout of SFA and SFB: https://github.com/flashinfer-ai/flashinfer/blob/main/include/flashinfer/gemm/gemm_groupwise_sm120.cuh#L122-L123, looks like the contiguity is not required (https://github.com/NVIDIA/cutlass/blob/main/examples/87_blackwell_geforce_gemm_blockwise/87b_blackwell_geforce_fp8_bf16_gemm_groupwise.cu#L304-L305).

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

looks like the contiguity is not required

I don't understand, if we allow non-contiguous SFA/SFB, at least we should pass the strides from tensors to kernels but I didn't notice the logic here.

CHECK_CONTIGUOUS(SFB);

DISPATCH_SCALE_MAJOR_K(scale_major_mode, SCALE_MAJOR_K, [&] {
return DISPATCH_DLPACK_INPUT_OUTPUT_DTYPE(A->dtype, C->dtype, c_type_in, c_type_out, [&] {
return DISPATCH_SCALE_GRANULARITY(
Expand Down